Data Modelling Approach¶

  • Step 1: Importing the Relevant Libraries

  • Step 2: Data Inspection and Cleaning

  • Step 3: Data Exploration

  • Step 4: RFM Modelling

  • Step 5: find out the customers who are 'champions', 'Potential customers' and 'need attention

  • Step 6: Data Visualization After RFM Analysis

  • Step 7: K-Means Clustering Technique

Step 1: Importing the Relevant Libraries¶

In [1]:
%matplotlib inline
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import datetime as dt
from matplotlib.gridspec import GridSpec
import seaborn as sns
import plotly.express as px
from sklearn.cluster import KMeans
from sklearn.preprocessing import StandardScaler as ss

Step 2: Data Inspection and Cleaning¶

In [2]:
data = pd.read_excel("C:/Users/AkshayPharande/Downloads/Round 1 Assignment/Round 1 Assignment/sales_data.xlsx")
data.head()
Out[2]:
CustomerID TOTAL_ORDERS REVENUE AVERAGE_ORDER_VALUE CARRIAGE_REVENUE AVERAGESHIPPING FIRST_ORDER_DATE LATEST_ORDER_DATE AVGDAYSBETWEENORDERS DAYSSINCELASTORDER ... WEEK3_DAY16_DAY23_REVENUE WEEK4_DAY24_DAY31_REVENUE TIME_0000_0600_ORDERS TIME_0601_1200_ORDERS TIME_1200_1800_ORDERS TIME_1801_2359_ORDERS TIME_0000_0600_REVENUE TIME_0601_1200_REVENUE TIME_1200_1800_REVENUE TIME_1801_2359_REVENUE
0 22 124 11986.54 96.67 529.59 4.27 2016-12-30 2021-10-24 14.19 1 ... 2592.18 2409.71 12 61 37 14 1067.92 5674.65 3916.53 1327.44
1 29 82 11025.96 134.46 97.92 1.19 2018-03-31 2021-10-24 15.89 1 ... 2807.66 4106.19 1 13 24 44 30.25 3143.69 2205.47 5646.55
2 83 43 7259.69 168.83 171.69 3.99 2017-11-30 2021-10-24 33.12 1 ... 713.94 2861.90 0 6 12 25 0.00 707.98 2560.91 3990.80
3 95 44 6992.27 158.92 92.82 2.11 2019-04-09 2021-10-24 21.11 1 ... 997.02 1260.08 0 10 13 21 0.00 1908.99 2025.05 3058.23
4 124 55 6263.44 113.88 179.04 3.26 2020-10-23 2021-10-24 6.65 1 ... 2725.66 768.46 6 9 27 13 501.34 1197.83 3004.77 1559.50

5 rows × 40 columns

In [3]:
data.shape
Out[3]:
(5000, 40)
  • We have 5000 rows and 40 columns
In [4]:
data.isna().sum()
Out[4]:
CustomerID                   0
TOTAL_ORDERS                 0
REVENUE                      0
AVERAGE_ORDER_VALUE          0
CARRIAGE_REVENUE             0
AVERAGESHIPPING              0
FIRST_ORDER_DATE             0
LATEST_ORDER_DATE            0
AVGDAYSBETWEENORDERS         0
DAYSSINCELASTORDER           0
MONDAY_ORDERS                0
TUESDAY_ORDERS               0
WEDNESDAY_ORDERS             0
THURSDAY_ORDERS              0
FRIDAY_ORDERS                0
SATURDAY_ORDERS              0
SUNDAY_ORDERS                0
MONDAY_REVENUE               0
TUESDAY_REVENUE              0
WEDNESDAY_REVENUE            0
THURSDAY_REVENUE             0
FRIDAY_REVENUE               0
SATURDAY_REVENUE             0
SUNDAY_REVENUE               0
WEEK1_DAY01_DAY07_ORDERS     0
WEEK2_DAY08_DAY15_ORDERS     0
WEEK3_DAY16_DAY23_ORDERS     0
WEEK4_DAY24_DAY31_ORDERS     0
WEEK1_DAY01_DAY07_REVENUE    0
WEEK2_DAY08_DAY15_REVENUE    0
WEEK3_DAY16_DAY23_REVENUE    0
WEEK4_DAY24_DAY31_REVENUE    0
TIME_0000_0600_ORDERS        0
TIME_0601_1200_ORDERS        0
TIME_1200_1800_ORDERS        0
TIME_1801_2359_ORDERS        0
TIME_0000_0600_REVENUE       0
TIME_0601_1200_REVENUE       0
TIME_1200_1800_REVENUE       0
TIME_1801_2359_REVENUE       0
dtype: int64
  • NO missing values present in data and this is a clean dataset.

Step 3: Data Exploration¶

In [5]:
data.describe()
Out[5]:
CustomerID TOTAL_ORDERS REVENUE AVERAGE_ORDER_VALUE CARRIAGE_REVENUE AVERAGESHIPPING AVGDAYSBETWEENORDERS DAYSSINCELASTORDER MONDAY_ORDERS TUESDAY_ORDERS ... WEEK3_DAY16_DAY23_REVENUE WEEK4_DAY24_DAY31_REVENUE TIME_0000_0600_ORDERS TIME_0601_1200_ORDERS TIME_1200_1800_ORDERS TIME_1801_2359_ORDERS TIME_0000_0600_REVENUE TIME_0601_1200_REVENUE TIME_1200_1800_REVENUE TIME_1801_2359_REVENUE
count 5000.000000 5000.00000 5000.000000 5000.000000 5000.000000 5000.000000 5000.000000 5000.000000 5000.000000 5000.00000 ... 5000.000000 5000.000000 5000.000000 5000.000000 5000.000000 5000.000000 5000.000000 5000.000000 5000.000000 5000.000000
mean 40709.227800 12.87040 1681.523840 136.537378 46.036376 3.592574 163.159618 87.420000 1.629000 1.75440 ... 421.826908 474.462852 1.028800 3.746200 4.434000 3.661400 131.062636 486.863868 584.731626 478.865710
std 49949.848017 12.67988 1998.618678 91.651569 47.879226 2.021360 259.699496 80.156513 2.236506 2.43394 ... 643.449120 617.579321 2.174331 4.700234 5.044793 4.581894 331.289349 789.029911 804.290026 743.244248
min 1.000000 1.00000 38.500000 10.680000 0.000000 0.000000 0.000000 1.000000 0.000000 0.00000 ... 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
25% 1687.500000 3.00000 315.097500 83.025000 9.980000 2.500000 21.670000 7.000000 0.000000 0.00000 ... 0.000000 80.000000 0.000000 1.000000 1.000000 1.000000 0.000000 35.000000 89.990000 1.000000
50% 13765.000000 8.00000 966.725000 113.160000 24.985000 3.660000 57.635000 68.000000 1.000000 1.00000 ... 194.990000 264.485000 0.000000 2.000000 3.000000 2.000000 0.000000 236.440000 299.840000 208.990000
75% 71891.500000 20.00000 2493.072500 160.272500 76.862500 4.790000 170.357500 171.250000 2.000000 3.00000 ... 604.085000 651.487500 1.000000 6.000000 7.000000 5.000000 131.282500 661.952500 797.032500 620.117500
max 277160.000000 156.00000 34847.400000 1578.880000 529.590000 35.990000 1409.500000 207.000000 19.000000 23.00000 ... 12946.220000 8274.970000 34.000000 61.000000 58.000000 64.000000 7700.240000 23832.770000 17749.010000 12995.020000

8 rows × 38 columns

In [6]:
AVG_Daily_Orders = [data[['MONDAY_ORDERS','TUESDAY_ORDERS','WEDNESDAY_ORDERS','THURSDAY_ORDERS','FRIDAY_ORDERS','SATURDAY_ORDERS','SUNDAY_ORDERS']].mean()]
AVG_Daily_Revenue = [data[['MONDAY_REVENUE','TUESDAY_REVENUE','WEDNESDAY_REVENUE','THURSDAY_REVENUE','FRIDAY_REVENUE','SATURDAY_REVENUE','SUNDAY_REVENUE']].mean()]
In [7]:
AVG_Daily_Orders, AVG_Daily_Revenue
Out[7]:
([MONDAY_ORDERS       1.6290
  TUESDAY_ORDERS      1.7544
  WEDNESDAY_ORDERS    1.7980
  THURSDAY_ORDERS     2.1340
  FRIDAY_ORDERS       1.9462
  SATURDAY_ORDERS     1.6834
  SUNDAY_ORDERS       1.9254
  dtype: float64],
 [MONDAY_REVENUE       215.208336
  TUESDAY_REVENUE      233.510430
  WEDNESDAY_REVENUE    235.689294
  THURSDAY_REVENUE     265.949796
  FRIDAY_REVENUE       250.580554
  SATURDAY_REVENUE     219.642100
  SUNDAY_REVENUE       260.943330
  dtype: float64])

According to the data above, Thursday, Friday, and Sunday have higher daily average customer orders than other days. Additionally, the average revenue generated on these days is higher than on other days..¶

In [8]:
AVG_Weekly_Orders = [data[['WEEK1_DAY01_DAY07_ORDERS','WEEK2_DAY08_DAY15_ORDERS','WEEK3_DAY16_DAY23_ORDERS','WEEK4_DAY24_DAY31_ORDERS']].mean()]
AVG_Weekely_Revenue = [data[['WEEK1_DAY01_DAY07_REVENUE','WEEK2_DAY08_DAY15_REVENUE','WEEK3_DAY16_DAY23_REVENUE','WEEK4_DAY24_DAY31_REVENUE']].mean()]
In [9]:
AVG_Weekly_Orders, AVG_Weekely_Revenue
Out[9]:
([WEEK1_DAY01_DAY07_ORDERS    2.9978
  WEEK2_DAY08_DAY15_ORDERS    3.0626
  WEEK3_DAY16_DAY23_ORDERS    3.2300
  WEEK4_DAY24_DAY31_ORDERS    3.5800
  dtype: float64],
 [WEEK1_DAY01_DAY07_REVENUE    378.638346
  WEEK2_DAY08_DAY15_REVENUE    406.595734
  WEEK3_DAY16_DAY23_REVENUE    421.826908
  WEEK4_DAY24_DAY31_REVENUE    474.462852
  dtype: float64])

According to the graph above, average revenue and orders increase from week 1 to week 4, and most customers prefer to buy products in the last week of the month.¶

Step 4: RFM Modeling Technique :¶

Here we will calculate the Recency, Frequency and Monetary for the customers and those are defined as ;

  • Recency : How much time has elapsed since a customer's last activity or transaction with the brand? i.e DAYSSINCELASTORDER
  • Frequency : How often has a customer transacted or interacted? i.e. TOTAL_ORDERS

  • Monetary : How much a customer has spent with the brand during a particular period of time? i.e.REVENUE

    Therefore, We need DAYSSINCELASTORDER, TOTAL_ORDERS and REVENUE columns to do RFM Modelling.

In [10]:
#RFM factors calculation:
#converting the names of the columns
RFM_data = data.rename(columns = {'DAYSSINCELASTORDER' : 'Recency',
                          'TOTAL_ORDERS' : "Frequency",
                          'REVENUE' : "Monetary"},inplace = False)
RFM_data.head()
Out[10]:
CustomerID Frequency Monetary AVERAGE_ORDER_VALUE CARRIAGE_REVENUE AVERAGESHIPPING FIRST_ORDER_DATE LATEST_ORDER_DATE AVGDAYSBETWEENORDERS Recency ... WEEK3_DAY16_DAY23_REVENUE WEEK4_DAY24_DAY31_REVENUE TIME_0000_0600_ORDERS TIME_0601_1200_ORDERS TIME_1200_1800_ORDERS TIME_1801_2359_ORDERS TIME_0000_0600_REVENUE TIME_0601_1200_REVENUE TIME_1200_1800_REVENUE TIME_1801_2359_REVENUE
0 22 124 11986.54 96.67 529.59 4.27 2016-12-30 2021-10-24 14.19 1 ... 2592.18 2409.71 12 61 37 14 1067.92 5674.65 3916.53 1327.44
1 29 82 11025.96 134.46 97.92 1.19 2018-03-31 2021-10-24 15.89 1 ... 2807.66 4106.19 1 13 24 44 30.25 3143.69 2205.47 5646.55
2 83 43 7259.69 168.83 171.69 3.99 2017-11-30 2021-10-24 33.12 1 ... 713.94 2861.90 0 6 12 25 0.00 707.98 2560.91 3990.80
3 95 44 6992.27 158.92 92.82 2.11 2019-04-09 2021-10-24 21.11 1 ... 997.02 1260.08 0 10 13 21 0.00 1908.99 2025.05 3058.23
4 124 55 6263.44 113.88 179.04 3.26 2020-10-23 2021-10-24 6.65 1 ... 2725.66 768.46 6 9 27 13 501.34 1197.83 3004.77 1559.50

5 rows × 40 columns

In [11]:
RFM_data_1 = RFM_data[['CustomerID','Recency','Frequency','Monetary']]
In [12]:
RFM_data_1.head()
Out[12]:
CustomerID Recency Frequency Monetary
0 22 1 124 11986.54
1 29 1 82 11025.96
2 83 1 43 7259.69
3 95 1 44 6992.27
4 124 1 55 6263.44
In [13]:
# RFM_data Description/ Summary
RFM_data_1.iloc[:,1:4].describe()
Out[13]:
Recency Frequency Monetary
count 5000.000000 5000.00000 5000.000000
mean 87.420000 12.87040 1681.523840
std 80.156513 12.67988 1998.618678
min 1.000000 1.00000 38.500000
25% 7.000000 3.00000 315.097500
50% 68.000000 8.00000 966.725000
75% 171.250000 20.00000 2493.072500
max 207.000000 156.00000 34847.400000
From above result, we can observe that average recency of the customers are 87 days (approx), an average customer are purchasing the product 13 times and spending an average 1681.52 unitprice.¶
In [14]:
#Visualizing the Recency, Frequency and Monetary distributions.
i = 0
fig = plt.figure(constrained_layout = True,figsize = (20,5))
gs = GridSpec(1, 3, figure=fig)    

col = ['red','blue','green']
for var in list(RFM_data_1.columns[1:4]):
    plt.subplot(gs[0,i])
    sns.distplot(RFM_data_1[var],color= col[i])
    plt.title('Skewness ' + ': ' + round(RFM_data_1[var].skew(),2).astype(str))
    i= i+1

C:\Users\AkshayPharande\Conda\lib\site-packages\seaborn\distributions.py:2619: FutureWarning: `distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).
  warnings.warn(msg, FutureWarning)
C:\Users\AkshayPharande\Conda\lib\site-packages\seaborn\distributions.py:2619: FutureWarning: `distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).
  warnings.warn(msg, FutureWarning)
C:\Users\AkshayPharande\Conda\lib\site-packages\seaborn\distributions.py:2619: FutureWarning: `distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).
  warnings.warn(msg, FutureWarning)
From above distribution plots we can observe that all of these three metrics are positively skewed or right skewed. And also as per skewness coefficient values indicating same.¶

Segmentation :¶

In [15]:
#Here, we will divide the data set into 4 parts based on the quantiles.
quantiles = RFM_data_1.drop('CustomerID',axis = 1).quantile(q = [0.33,0.67])
quantiles.to_dict()
Out[15]:
{'Recency': {0.33: 19.0, 0.67: 128.0},
 'Frequency': {0.33: 4.0, 0.67: 17.0},
 'Monetary': {0.33: 427.09400000000005, 0.67: 1905.1959}}
In [16]:
#Creating the R,F and M scoring/segement function
#[1] Recency scoring (Negative Impact : Higher the value, less valuable)
def R_score(var,p,d):
    if var <= d[p][0.33]:
        return 1
    elif var <= d[p][0.67]:
        return 2
    else:
        return 3
#[2] Frequency and Monetary (Positive Impact : Higher the value, better the customer)
def FM_score(var,p,d):
    if var <= d[p][0.33]:
        return 3
    elif var <= d[p][0.67]:
        return 2
    else:
        return 1


#Scoring:
RFM_data_1['R_score'] = RFM_data_1['Recency'].apply(R_score,args = ('Recency',quantiles,))
RFM_data_1['F_score'] = RFM_data_1['Frequency'].apply(FM_score,args = ('Frequency',quantiles,))
RFM_data_1['M_score'] = RFM_data_1['Monetary'].apply(FM_score,args = ('Monetary',quantiles,))
RFM_data_1.head()

C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\612544624.py:21: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  RFM_data_1['R_score'] = RFM_data_1['Recency'].apply(R_score,args = ('Recency',quantiles,))
C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\612544624.py:22: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  RFM_data_1['F_score'] = RFM_data_1['Frequency'].apply(FM_score,args = ('Frequency',quantiles,))
C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\612544624.py:23: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  RFM_data_1['M_score'] = RFM_data_1['Monetary'].apply(FM_score,args = ('Monetary',quantiles,))
Out[16]:
CustomerID Recency Frequency Monetary R_score F_score M_score
0 22 1 124 11986.54 1 1 1
1 29 1 82 11025.96 1 1 1
2 83 1 43 7259.69 1 1 1
3 95 1 44 6992.27 1 1 1
4 124 1 55 6263.44 1 1 1
In [17]:
#Now we will create : RFMGroup and RFMScore
RFM_data_1['RFM_Group'] = RFM_data_1['R_score'].astype(str) + RFM_data_1['F_score'].astype(str) + RFM_data_1['M_score'].astype(str)

#Score
RFM_data_1['RFM_Score'] = RFM_data_1[['R_score','F_score','M_score']].sum(axis = 1)
RFM_data_1.head()

C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\1189733458.py:2: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  RFM_data_1['RFM_Group'] = RFM_data_1['R_score'].astype(str) + RFM_data_1['F_score'].astype(str) + RFM_data_1['M_score'].astype(str)
C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\1189733458.py:5: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  RFM_data_1['RFM_Score'] = RFM_data_1[['R_score','F_score','M_score']].sum(axis = 1)
Out[17]:
CustomerID Recency Frequency Monetary R_score F_score M_score RFM_Group RFM_Score
0 22 1 124 11986.54 1 1 1 111 3
1 29 1 82 11025.96 1 1 1 111 3
2 83 1 43 7259.69 1 1 1 111 3
3 95 1 44 6992.27 1 1 1 111 3
4 124 1 55 6263.44 1 1 1 111 3

RFM Scores have been calculated now we will use this score to make segments of the customers and define level of loyality.

Step 5: find out the customers who are 'champions', 'Potential customers' and 'need attention'¶

In [18]:
#Creating the Customer segments/ Loyality_level
Segment = ['champions','Potential customers','need attention']
cuts = pd.qcut(RFM_data_1['RFM_Score'],q = 3,labels=Segment)
RFM_data_1['Segment'] = cuts.values
RFM_data_1.tail(15)

C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\3162852954.py:4: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  RFM_data_1['Segment'] = cuts.values
Out[18]:
CustomerID Recency Frequency Monetary R_score F_score M_score RFM_Group RFM_Score Segment
4985 173166 207 3 118.68 3 3 3 333 9 need attention
4986 173176 207 4 118.59 3 3 3 333 9 need attention
4987 173219 207 1 118.49 3 3 3 333 9 need attention
4988 173315 207 1 118.49 3 3 3 333 9 need attention
4989 173503 207 2 118.08 3 3 3 333 9 need attention
4990 173766 207 4 117.79 3 3 3 333 9 need attention
4991 173792 207 1 117.59 3 3 3 333 9 need attention
4992 173842 207 1 117.49 3 3 3 333 9 need attention
4993 173857 207 1 117.49 3 3 3 333 9 need attention
4994 173944 207 1 117.49 3 3 3 333 9 need attention
4995 173946 207 1 117.49 3 3 3 333 9 need attention
4996 173987 207 1 117.49 3 3 3 333 9 need attention
4997 174004 207 1 117.49 3 3 3 333 9 need attention
4998 174038 207 1 117.49 3 3 3 333 9 need attention
4999 200783 207 2 94.14 3 3 3 333 9 need attention

We have classified our customer into four segments based on their R,F and M scores.

Step 6: Data Visualization¶

_Visualization for Recency, Frequency and Monetary : RFMSegment ¶

1.Recency V/s Frequency¶

In [19]:
fig = px.scatter(RFM_data_1,x = "Recency", y = "Frequency",color = "Segment")
fig.show()

We can see the customers whose Recency is less than two month have high Frequency i.e the customers buying more when their recency is less.¶

2.Frequency V/s Monetary¶

In [20]:
fig = px.scatter(RFM_data_1,x = "Monetary", y = "Frequency",color = "Segment")
fig.show()

We can see, customers buying frequently are spending less amount.¶

3. Recency V/s Monetary¶

In [21]:
fig = px.scatter(RFM_data_1,x = "Recency", y = "Monetary",color = "Segment")
fig.show()

We can see the customers whose Recency is less than two month have high Monetary i.e the customers spending more when their recency is less.¶

4. RFM_Segment¶

In [22]:
plt.figure(figsize=(10,7))
sns.countplot('Segment',data=RFM_data_1,palette='ocean')

C:\Users\AkshayPharande\Conda\lib\site-packages\seaborn\_decorators.py:36: FutureWarning:

Pass the following variable as a keyword arg: x. From version 0.12, the only valid positional argument will be `data`, and passing other arguments without an explicit keyword will result in an error or misinterpretation.
Out[22]:
<AxesSubplot:xlabel='Segment', ylabel='count'>

From Above chart It is clear that almost 50% customers are spend time and buy products and around 25-30% customers are potential customers and 20-25% customers need attention.¶

Step 7: K-Means Clustering Technique :¶

How the K-means algorithm works ?¶

Clustering is the process of dividing the entire data into groups (also known as clusters) based on the patterns in the data.

To process the learning data, the K-means algorithm in data mining starts with a first group of randomly selected centroids, which are used as the beginning points for every cluster, and then performs iterative (repetitive) calculations to optimize the positions of the centroids.

To create the customer segementation based on the K-Means Clustering based on the R, F, and M Scores: Before that we will bring them into same scale and normalise them.

In [23]:
# First will focus on the negativ and zero before the transformation.
def right_treat(var):
    if var <= 0:
        return 1
    else:
        return var

# Describing the data
RFM_data_1.describe()
Out[23]:
CustomerID Recency Frequency Monetary R_score F_score M_score RFM_Score
count 5000.000000 5000.000000 5000.00000 5000.000000 5000.000000 5000.000000 5000.000000 5000.000000
mean 40709.227800 87.420000 12.87040 1681.523840 1.995400 2.047200 2.000000 6.042600
std 49949.848017 80.156513 12.67988 1998.618678 0.812595 0.817744 0.812485 1.880929
min 1.000000 1.000000 1.00000 38.500000 1.000000 1.000000 1.000000 3.000000
25% 1687.500000 7.000000 3.00000 315.097500 1.000000 1.000000 1.000000 4.000000
50% 13765.000000 68.000000 8.00000 966.725000 2.000000 2.000000 2.000000 6.000000
75% 71891.500000 171.250000 20.00000 2493.072500 3.000000 3.000000 3.000000 7.000000
max 277160.000000 207.000000 156.00000 34847.400000 3.000000 3.000000 3.000000 9.000000

From above we can see that there is no 0 values present in Frequency , Monetary and Recency.

In [24]:
#Applying on the data.
RFM_data_1['Recency'] = RFM_data_1['Recency'].apply(lambda x : right_treat(x))
RFM_data_1['Frequency'] = RFM_data_1['Frequency'].apply(lambda x : right_treat(x))
RFM_data_1['Monetary'] = RFM_data_1['Monetary'].apply(lambda x : right_treat(x))

#Checking the Skewness of R, F and M
print('Recency Skewness : ' + RFM_data_1['Recency'].skew().astype(str))
print('Frequency Skewness : ' + RFM_data_1['Frequency'].skew().astype(str))
print('Monetary Skewness : ' + RFM_data_1['Monetary'].skew().astype(str))

Recency Skewness : 0.390002913835805
Frequency Skewness : 2.205765054664766
Monetary Skewness : 3.9057746675131164

C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\3514597734.py:2: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\3514597734.py:3: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\3514597734.py:4: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

All the variables are right skewed, so will make log transformation of it.

In [25]:
#Log Transformation
log_RFM_data = RFM_data_1[['Recency','Frequency','Monetary']].apply(np.log,axis = 1).round(3)
In [26]:
i = 0
fig = plt.figure(constrained_layout = True,figsize = (20,5))
gs = GridSpec(1, 3, figure=fig)    

col = ['red','blue','green']
for var in list(log_RFM_data.columns[0:3]):
    plt.subplot(gs[0,i])
    sns.distplot(log_RFM_data[var],color= col[i])
    plt.title('Skewness ' + ': ' + round(log_RFM_data[var].skew(),2).astype(str))
    i= i+1
log_RFM_data.describe()

C:\Users\AkshayPharande\Conda\lib\site-packages\seaborn\distributions.py:2619: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).

C:\Users\AkshayPharande\Conda\lib\site-packages\seaborn\distributions.py:2619: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).

C:\Users\AkshayPharande\Conda\lib\site-packages\seaborn\distributions.py:2619: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).
Out[26]:
Recency Frequency Monetary
count 5000.000000 5000.000000 5000.000000
mean 3.380851 2.041230 6.807619
std 2.008518 1.093141 1.173694
min 0.000000 0.000000 3.651000
25% 1.946000 1.099000 5.753000
50% 4.220000 2.079000 6.874000
75% 5.143250 2.996000 7.821250
max 5.333000 5.050000 10.459000

Therefore all the variables are now approximately normally distributed. Will make all of them on the same scale as Monetary is little large in values.

In [27]:
#Scaling the data
from sklearn.preprocessing import StandardScaler
ss = StandardScaler()
Scaled_RFM_data = ss.fit_transform(log_RFM_data)
Scaled_RFM_data = pd.DataFrame(Scaled_RFM_data,columns=log_RFM_data.columns,index=log_RFM_data.index)
In [28]:
# Will search the optimal number of cluster based on the Elbow Method as below:
SS_distance = {}
for k in range(1,20):
    mod = KMeans(n_clusters= k, max_iter=1000,init = 'k-means++')
    mod = mod.fit(Scaled_RFM_data)
    SS_distance[k] = mod.inertia_

#Plotting the sum of square distance values and numbers of clusters
plt.figure(figsize = (15,5))
sns.pointplot(x = list(SS_distance.keys()), y = list(SS_distance.values()))
plt.xlabel("Number of clusters")
plt.ylabel("Sum of square Distances")
plt.title("Elbow Techinque to find the optimal cluster size")
Out[28]:
Text(0.5, 1.0, 'Elbow Techinque to find the optimal cluster size')

We can observe that as the number of cluster increases the sum of square distance are becoming lesser. And will take the count of cluster where this elbow is bending. In our cases, sum of square distance is dramatically decreasing at K = 3, so this is optimal value to choose for no of clusters.

In [29]:
# Now we will perform K- means clustering on the data set.
KM_clust = KMeans(n_clusters= 3, init = 'k-means++',max_iter = 1000)
KM_clust.fit(Scaled_RFM_data)

# Mapping on the data
RFM_data_1['Cluster'] = KM_clust.labels_
RFM_data_1['Cluster'] = 'Cluster' + RFM_data_1['Cluster'].astype(str)
RFM_data_1.head()

C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\821865125.py:6: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

C:\Users\AkshayPharande\AppData\Local\Temp\ipykernel_3856\821865125.py:7: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
Out[29]:
CustomerID Recency Frequency Monetary R_score F_score M_score RFM_Group RFM_Score Segment Cluster
0 22 1 124 11986.54 1 1 1 111 3 champions Cluster1
1 29 1 82 11025.96 1 1 1 111 3 champions Cluster1
2 83 1 43 7259.69 1 1 1 111 3 champions Cluster1
3 95 1 44 6992.27 1 1 1 111 3 champions Cluster1
4 124 1 55 6263.44 1 1 1 111 3 champions Cluster1

Clusters have been created based on the values of recency, frequency and monetary with the help of K-Means Clustering.

Visualization for Recency, Frequency and Monetary : Cluster Groups¶

In [30]:
# Recency V/s Frequency
fig = px.scatter(RFM_data_1,x = 'Recency',y = 'Frequency', color = 'Cluster')
fig.show()

# Frequency V/s Monetary
fig = px.scatter(RFM_data_1,x = 'Monetary',y = 'Frequency', color = 'Cluster')
fig.show()

# Recency V/s Monetary
fig = px.scatter(RFM_data_1,x = 'Recency',y = 'Monetary', color = 'Cluster')
fig.show()
  • Plotting Frequency and Recency: here, the blue group frequently purchases products and they are the most recency one. The green ones tried to purchase recently but they are not frequent buyers which we can determine that they are the new customers.
  • Plotting Frequency and Monetary: even here the blue group of customers tried to purchase more and frequently whereas the green group is very little frequency and spends very little.
  • Plotting Recency and monetary: from the graph we can say that the blue group the ones who like to spend more and they are the recent customers.

Step 8: Findings¶

  • 1: Most of the customer orders placed during last week of the month
  • 2: Thursday, Friday, and Sunday are the most prominent days for customers to purchase products.
  • 3: average recency of the customers are 87 days (approx), an average customer are purchasing the product 13 times and spending an average 1681.52 unitprice
  • 4: The customers whose Recency is less than two month have high Frequency i.e the customers buying more when their recency is less
  • 5: customers buying frequently are spending less amount.
  • 6: the customers whose Recency is less than two month have high Monetory i.e the customers spending more when their recency is less.
  • 7: almost 50% customers are spend time and buy products and around 25-30% customers are potential customers and 20-25% customers need attention.

Step 8: Suggestions¶

I. Based on the above R-F-M score, we can give some Recommendations.

  • 1: Champions: We can Reward them for their multiples purchases. They can be early adopters to very new products. Suggest them “Refer a friend”. Also, they can be the most loyal customers that have the habit to order.
  • 2: Potential customers: Create loyalty cards in which they can gain points each time of purchasing and these points could transfer into a discount
  • 3: Need attention: Send them personalized emails/messages/notifications to encourage them to order. Also Notify them about the discounts to keep them spending more and more money on your products.

II. Based On the visualization of data :

  • 1. Allow discounts on products at the beginning of the month so that customers can start spending from week one..
  • 2. Take survey from need attention customers about service, offers and what they required from us.

THANK YOU...¶

In [ ]: